Two basic features of the ACS data products raised key questions for the panel’s consideration: (1) the plan to use multiyear estimates from aggregating monthly samples from the ACS to replace point-in-time estimates from the decennial census long-form sample and (2) the plan to use census-based population estimates at the level of counties (or groups of small counties) for July 1 of each year to calibrate the sample-based ACS estimates. The purpose of using the population estimates, which are developed from the previous census updated with births, deaths, and migrant flows from administrative records, as controls is to reduce the effects of sampling error and compensate for any incompleteness of coverage in the population surveyed in the ACS. There are also separate controls for housing units.
The use of multiyear averages is central to the ACS design. The question it raises is the extent to which users can easily apply the ACS data products to the important and varied uses that, until now, were met through the long-form-sample data products for the census year. What does it mean to have a 5-year estimate for an area of, say, the poverty level or the average length of time to commute to work cumulated from 60 months’ worth of data? How should users interpret differences in estimates for the same geographic area, such as a large county, that will be available from data cumulated over 1, 3, and 5 years? Turning to the detailed procedures for producing multiyear estimates, does the Census Bureau’s plan to weight the data to the average population over the period represent the optimal approach, or are there other approaches that might on balance have better properties?
The decision to use controls for counties (or groups of small counties) for specific population groups (defined by age, sex, race, and ethnicity) for July 1 each year raises several questions. One question concerns the effect of controlling the ACS monthly samples spread over a year to point-in-time population estimates that are updated from April 1 population counts from the last census. Another question concerns whether the county population estimates by age, sex, race, and ethnicity are of adequate quality to be used for this purpose. Might, for instance, controlling to population estimates at higher levels of geographic aggregation or using fewer population groups as controls offer advantages over the current plan?
In addition to these two primary areas of investigation, the panel needed to examine the ACS as a data collection and production system to answer questions concerning the functionality and usability of ACS data products. This examination led the panel to address four areas of ACS operations: (1) sampling for housing units, including initial sampling from the MAF and subsampling of nonrespondents for follow-up by computer-assisted personal interviewing; (2) data collection for housing units,